Light-Weight Ontologies for Scrutable User Modelling

This thesis is concerned with the ways light-weight ontologies can support scrutability for large user models and the user modelling process. It explores the role that light-weight ontologies can play, and how they can be exploited, for the purpose of creating and maintaining large, scrutable user models consisting of hundreds of components. We address problems in four key areas: ontology creation, metadata annotation, creation and maintenance of large user models, and user model visualisation, with a goal to provide a simple and adaptable approach that maintains scrutability. Each of these key areas presents a number of challenges that we address. Our solution is the development of a toolkit, LOSUM, which consists of a number of tools to support the user modelling process. It incorporates light-weight ontologies to fulfill a number of roles: aiding in metadata creation, providing structure for large user model visualisation, and as a means to reason across granularities in the user model. In conjunction with this, LOSUM also features a novel visualisation tool, SIV, which performs a dual role of ontology and user model visualisation, supporting the process of ontology creation, metadata annotation, and user model visualisation. We evaluated our approach at each stage with small user studies, and conducted a large scale integrative evaluation of these approaches together in an authentic learning context with 114 students, of whom 77 had exposure to their learner models through SIV. The results showed that students could use the interface and understand the process of user model construction. The flexibility and adaptability of the toolkit has also been demonstrated in its deployment in several other application areas.